Missing virtio/scsi modules for arm64?

[Steve McIntyre <steve@einval.com>]

Hey folks,

I;m just trying to do a buster install of arm64 (using buster d-i RC1)
in a qemu VM, and it's failing to find the virtio cdrom. Checking into
this, it looks like we're missing some modules in the installer
environment. Going back to alpha 5 (kernel 4.19.12-1), we get the
following modules loaded:

Module                  Size  Used by
efivars                20480  0
nls_utf8               16384  1
isofs                  49152  1
sr_mod                 32768  1
cdrom                  61440  1 sr_mod
virtio_scsi            20480  1
scsi_mod              233472  2 virtio_scsi,sr_mod
virtio_net             49152  0
virtio_blk             20480  0
net_failover           20480  1 virtio_net
failover               16384  1 net_failover
gpio_keys              20480  0
virtio_pci             28672  0
virtio_mmio            20480  0
virtio_ring            28672  5 virtio_mmio,virtio_scsi,virtio_pci,virtio_blk,virtio_net
virtio                 16384  5 virtio_mmio,virtio_scsi,virtio_pci,virtio_blk,virtio_net

In RC1 (4.19.28-2), we only get:
Module                  Size  Used by                                           
gpio_keys              20480  0                                                 
virtio_pci             28672  0                                                 
virtio_mmio            20480  0                                                 
virtio_ring            28672  2 virtio_mmio,virtio_pci                          
virtio                 16384  2 virtio_mmio,virtio_pci                          

and if I look for things like virtio_scsi or sr_mod they're just not
in the initramfs any more which will explain what's going on. Ben, I
can see that in d-i you've changed the package lists recently (i.e. in
between the 2 d-i releases):

commit 5b3bcf67a7d8ba7745612a449cede24afeb97015
Author: Ben Hutchings <ben@decadent.org.uk>
Date:   Tue Feb 12 21:18:40 2019 +0000

    build/pkg-lists: Make {hyperv,virtio}-modules packages optional
    
    I intend to remove these udebs in a later upload of linux, moving
    the drivers into per-driver-class packages.

and I'm guessing these changes are related to what I'm seeing. Can you
suggest what we should be doing to fix things up, please? I'm at a
total loss to see where amd64 is gettting d-i modules from nowadays,
and arm64 should be reasonably similar in terms of things like block
device support modules here.

Cheers,

Steve

-- 
Steve McIntyre, Cambridge, UK.                                steve@einval.com
"When C++ is your hammer, everything looks like a thumb." -- Steven M. Haflich